Multiscale Human Liver Volume Electron Microscopy

Introduction

Using a deep learning-based segmentation framework, we generated comprehensive labels across vascular, cellular, and subcellular levels, enabling quantitative analysis of bile duct–cholangiocyte organization and sinusoidal branch geometry. At the organelle scale, analysis of 35,790 mitochondria revealed distinct morphological profiles and spatial distributions. Examination of mitochondrial–endoplasmic reticulum (ER) spatial relationships uncovered characteristic ER-associated mitochondrial narrowing, indicative of fission and fusion activity.

Installation

1. Create a virtual environment to install the required packages. This takes less than 1 min. An example setup script is provided in create_run_example.slurm.

2. Clone the repository:

   git clone https://github.com/BaderLab/Multiscale_human_liver_vEM.git

3. Install dependencies:

   cd Multiscale_human_liver_vEM
   pip install -r requirements.txt

   The requirements.txt includes packages needed for both nnUNet and SAM2.

Getting Started

1. Vascular and Cellular Level Segmentation

We provide a script that uses SAM2 to generate 3D instance masks from input prompts (GPU is required and we used H100 GPU when running this):

python sam2maskpropagator.py

2. Organelle Segmentation

Organelle segmentation was performed using nnUNet with pretraining and fine-tuning (GPU is required and we used H100 GPU when running this). Trained model checkpoints for all segmented organelles are available on Zenodo.

3. Mitochondrial Morphology Feature Extraction

After obtaining organelle masks, morphological features of mitochondria were extracted using PyRadiomics:

python morphology_features.py

4. Mitochondria–ER Interaction Analysis

To analyze mitochondria–ER spatial interactions:

python mito_er_analysis.py

Acknowledgements

We thank the SAM2 and nnUNet teams for making their source code publicly available. We also thank the PyRadiomics team for their open-source morphological feature extraction package. We gratefully acknowledge OpenOrganelle and Parlakgül et al. (2022) for making the mouse liver volume electron microscopy data publicly available.

Citation

@article{multiscale_human_liver,
  title   = {},
  author  = {},
  journal = {},
  volume  = {},
  pages   = {},
  year    = {}
}